I was in the same boat you were about 4 days ago. I found that using picamera 2 was the only way I could get it to work.

from picamera2 import Picamera2
from ultralytics import YOLO
import cv2
import numpy as np

# Load the YOLO model


# Initialize the Raspberry Pi camera
picam2 = Picamera2()
config = picam2.create_preview_configuration(main={"size": (640, 480)})
picam2.configure(config)
picam2.start()

# Function to process frames with YOLO
def process_frame(frame):

# Display video feed
try:
    print("Press 'q' to quit.")
    while True:
        # Capture frame from the Raspberry Pi camera
        frame = picam2.capture_array()
        frame = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)  # Convert to OpenCV's BGR format

        # Process frame with YOLO
        processed_frame = process_frame(frame)

        # Show the video feed
        cv2.imshow("Video Feed", processed_frame)

        # Exit if 'q' is pressed
        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
finally:
    picam2.stop()
    cv2.destroyAllWindows()

The problem I am trying to fix is to display the video feed on the display so that I can see the bounding boxes on the objects YOLO detected. But cv2.imshow does not seem to work for some reason. The other parts of the code work tho.